Chatbot Statistics 2026: 52% of Execs Run AI Agents, 74% See ROI

AI chatbot statistics for 2026 show fast business adoption and cautious customers. 52% of executives say their organizations run AI agents in production and 74% report ROI within a year (Google Cloud). US adults name customers preferring a person as the top obstacle to AI at work (Anthropic). Dante AI re-checked every figure against its source on 13 November 2026.

How we built this page: every number below was read again on 13 November 2026 from the publisher’s own page, linked next to the figure, with its publication date. We dropped any figure we could not trace to a primary source, so this list is shorter than most. Dante AI’s own company figures are not included.

What do AI chatbot statistics show in 2026?

They show businesses moving faster than customers. In Google Cloud’s 2025 study, 52% of executives say their organizations run AI agents in production, 39% have more than ten, and 74% report ROI within a year. In Anthropic’s 2026 survey of US adults, customers preferring a person was the obstacle to AI at work chosen most often.

The detail behind those numbers. On adoption, Google Cloud’s 2025 ROI of AI study, based on a survey of 3,466 senior leaders of global enterprises, found that 52% of executives say their organizations deploy AI agents in production, and 39% have more than ten. On return, 74% of executives report ROI within the first year. On customer expectation, a nationally representative survey of 10,980 US adults, run by Morning Consult for Anthropic in August 2026, found that “customers preferring to deal with a person” was the obstacle to AI at work chosen most often, ahead of privacy and reliability.

Put together: businesses are deploying, the money case is landing quickly, and the open question is how customers feel about talking to an AI agent. That is why the practical answer is an AI agent that resolves the routine questions and hands the rest to a person.

What are the key chatbot stats, with sources? (table)

The table lists all 13 chatbot stats on this page. Each row gives the statistic, the publisher it comes from with a link to the page we read, and the year it was published. We re-read every source on 13 November 2026 and dropped any figure we could not open and confirm, so no row is an estimate.

Two caveats apply to the whole table. The Anthropic figures describe how people and businesses use one AI model family, Claude, so treat them as a strong signal rather than a market total. Google Cloud sells AI agent platforms, and its survey covers large global enterprises, not small businesses.

How many businesses use AI chatbots and AI agents?

Among large global enterprises, most already do. 52% of executives in Google Cloud’s survey of 3,466 senior leaders say their organizations deploy AI agents in production, and 39% say they have deployed more than ten. Among US adults, about one in five use AI at work regularly, so everyday use is still a minority habit.

Agents have moved from pilots to production in large companies. In Google Cloud, The ROI of AI, 4 Sep 2025, 52% of executives say their organizations deploy AI agents in production and 39% say they have deployed more than ten across the enterprise. Google Cloud names customer service resolution as one of the first use cases to start with, because it gives a clear ROI metric.

Inside one model provider’s business traffic, the pattern is delegation. Anthropic Economic Index, 15 Sep 2025 found that 77% of business API uses follow automation patterns, against about 50% for people chatting on Claude.ai. By November 2025, office and administrative support work had grown to 13% of business API transcripts, up 3 points in three months (Anthropic Economic Index, 15 Jan 2026).

Among the public, use at work is still a minority habit: about one in five US adults surveyed use AI at work regularly, and US adults expect 42% of the work tasks AI could do to be done with AI by 2030 (Anthropic, The AI expectations gap, 16 Oct 2026).

Do customers prefer AI chatbots or a person?

Many still want a person within reach. In Anthropic’s survey of 10,980 US adults, customers preferring to deal with a person was the obstacle to AI at work chosen most often. 42% of adults who do not use AI regularly at work put it in their top three, and 37% of regular users did.

When Anthropic, The AI expectations gap, 16 Oct 2026 asked US adults to pick the biggest obstacles to AI being used at work in 2030, “customers preferring to deal with a person” was chosen most often, followed by data privacy and security, AI not being reliable enough, and laws or liability. 42% of adults who do not use AI regularly at work put it in their top three, and 37% of regular users did.

The lesson for a support team is not “avoid AI”. It is: let the AI agent answer the questions it can answer well, and make the route to a person obvious and fast. Our comparison of an AI chatbot and live chat covers how to split the work, and AI chatbot response time explains why speed is the first thing customers notice.

How much customer service work does AI resolve on its own?

No public average is reliable, because vendors count a resolution in different ways. The clearest primary data is on delegation: in Anthropic’s business API data, 97% of economic tasks show automation-dominant patterns, and customer service tasks such as billing support are prevalent. The resolution rate that matters is the one you measure on your own questions.

Anthropic Economic Index, 15 Sep 2025 found that 97% of economic tasks seen in business API use show automation-dominant patterns, against 47% on Claude.ai. On Claude.ai itself, conversations where the user hands over the whole task rose from 27% in January 2025 to 39% in August 2025, then eased to 32% in November 2025 (Anthropic Economic Index, 15 Jan 2026).

Customer service shows up directly in that business data. Anthropic Economic Index, 24 Mar 2026 notes that customer service tasks, for example automated support for payment and billing issues, are prevalent in its API data, which gives Customer Service Representatives a high observed exposure to automation. It does not publish a share for them.

Vendor resolution rates are usually self-reported and measured in different ways (resolved per conversation, per ticket, with or without a follow-up survey), so we left them out. The resolution rate that matters is yours, on your own questions, and the worked example below shows how to measure it in a week.

What ROI do AI chatbots deliver?

In large companies, the return comes quickly. Google Cloud reports that 74% of executives achieve ROI from AI within the first year, and that 39% of those who report productivity gains have seen productivity at least double. For a small team, compare what each answered question costs today with what an AI agent costs.

The source: Google Cloud, The ROI of AI, 4 Sep 2025 reports that 74% of executives achieve ROI within the first year, and that 39% of those who report productivity gains have seen productivity at least double.

For a small team the cost side is simpler: what you pay per answered question today against what the AI agent costs. Our AI chatbot ROI guide walks through the full calculation.

Are there reliable AI chatbot traffic statistics?

Not from a source we could check. Visitor and referral numbers for chatbot sites are widely quoted, but they come from web analytics estimates we could not open and confirm on 13 November 2026, so none appear here. If you need traffic data, use your own analytics: count the visitors who open your chat and the questions they ask.

That count is also the starting point for the worked example below. Adding an AI agent to your website lets you see how many visitors actually use chat, rather than relying on a market estimate.

How do you work out your own chatbot numbers? A worked example

Work out four numbers from one month of your own support data: cost per answered question, the share of questions that repeat, how many of the top 20 repeats an AI agent answers correctly, and the hours that frees. The example below uses an invented homeware shop, so every figure in it is made up.

The business: an online homeware shop with two support staff.

1. Cost per answered question. One month of fully loaded support cost is $4,800 and the team handled 1,200 questions, so each answer costs $4.00. (Use your own figures; these are made up.)
2. Repeat share. Export the month’s tickets and chats and group them by question. In our example the 20 most repeated questions (delivery times, returns, sizing, order status) make up 600 of the 1,200, or 50%.
3. Resolution test. Train an AI agent on the website, the returns policy and the product sheets, then ask it those 20 questions. Score each one right, partly right or wrong. Say 17 of 20 are right: an 85% pass rate on the repeat questions.
4. Estimate. 85% of 600 repeat questions is about 510 questions a month the AI agent could take, roughly $2,040 of support time at $4.00 each. Set that against the plan price on the pricing page.
5. Keep the person in the loop. Route the other questions, and anyone who asks, to the team. Given the customer-expectation data above, this is what keeps satisfaction up while the numbers improve.

Dante AI’s free plan gives you 100 credits a month, which is enough to run this 20-question test properly before you spend anything. Adding the AI agent to your website takes one script tag once the test passes.

Which chatbot statistics should you be careful with?

Be careful with any number you cannot open at its source. Many widely shared chatbot statistics come from roundups that quote other roundups, with no survey, sample size or date. Market forecasts, vendor resolution rates and enterprise-only samples are also easy to misread. The four checks below take a minute per figure.

• Numbers with no named source or date. Many widely shared chatbot statistics trace back to old surveys or to other roundups. If you cannot open the original, do not quote it.
• Market forecasts. Market-size figures vary several times over between research firms because each defines “chatbot” differently.
• Vendor resolution rates. Useful as a ceiling, but each vendor counts a resolution its own way.
• Enterprise-only samples. Surveys of large companies, like Google Cloud’s, describe what big teams see. Small businesses usually move faster and with less integration work.

For a wider view of where AI fits in a support team, see the ultimate guide to AI in customer service.

Where do these chatbot statistics come from?

Every figure comes from six primary pages: two from Google Cloud and four from Anthropic’s research site, listed below with their publication dates. We fetched each page on 13 November 2026 and found every number on this page in its text. Dante AI’s own usage figures are not included.

• Google Cloud, “The ROI of AI: How agents are delivering for business”, Oliver Parker, published 4 September 2025: cloud.google.com. Survey size from Google Cloud’s ROI of AI in CX report page (3,466 senior leaders of global enterprises).
• Anthropic, “The AI expectations gap”, published 16 October 2026; US adults surveyed by Morning Consult, 11 to 23 August 2026, 10,980 respondents, weighted to be nationally representative: anthropic.com.
• Anthropic Economic Index report, “Uneven geographic and enterprise AI adoption”, published 15 September 2025: anthropic.com.
• Anthropic Economic Index report, “Economic primitives”, published 15 January 2026: anthropic.com.
• Anthropic Economic Index report, “Learning curves”, published 24 March 2026: anthropic.com.